rbinrs/Qwen2.5-Coder-1.5B-Instruct-abliterated
The rbinrs/Qwen2.5-Coder-1.5B-Instruct-abliterated model is a 1.5 billion parameter instruction-tuned causal language model, derived from the Qwen2.5-Coder architecture. Developed by huihui-ai, this version has been uncensored using the abliteration technique, making it suitable for applications requiring less restrictive content generation. It maintains a 32768 token context length and is primarily designed for code-related tasks, offering an uncensored alternative to the original Qwen2.5-Coder-1.5B-Instruct.
Loading preview...
Model Overview
This model, rbinrs/Qwen2.5-Coder-1.5B-Instruct-abliterated, is a 1.5 billion parameter instruction-tuned causal language model based on the Qwen2.5-Coder architecture. It was created by huihui-ai as an uncensored version of the original Qwen2.5-Coder-1.5B-Instruct, utilizing the abliteration technique. This process aims to remove content restrictions, offering greater flexibility in generated output.
Key Characteristics
- Uncensored Version: Modified using the abliteration technique to provide less restricted content generation compared to its base model.
- Coder-focused: Inherits the code generation capabilities of the Qwen2.5-Coder series.
- Parameter Count: A compact 1.5 billion parameters, making it suitable for resource-constrained environments.
- Context Length: Supports a substantial context window of 32768 tokens.
- Availability: Part of a family of uncensored Qwen2.5-Coder models available in various sizes (0.5B, 1.5B, 3B, 7B, 14B, 32B).
Performance Insights
Evaluations show that the abliterated version generally maintains competitive performance, and in some cases, slightly improves on specific benchmarks compared to the original Qwen2.5-Coder-1.5B-Instruct:
- IF_Eval: Achieves 45.41, a slight improvement over the original's 43.43.
- MMLU Pro: Scores 20.57.
- TruthfulQA: Scores 41.9.
- BBH: Scores 36.09.
- GPQA: Scores 26.13.
Usage
This model can be easily integrated into applications using the Hugging Face transformers library. It also has an official Ollama integration, allowing for local deployment via ollama run huihui_ai/qwen2.5-coder-abliterate:1.5b.